The establishment of associations between structure and function among various areas of the brain is an important step in the identification of neurologic mechanisms. The pursuit of this goal requires the geometrical co-registration of digital images in two types of major applications: (1) data fusion, in the case brain images from the same subject were acquired by different modalities; (2) data comparison, for the detection of significant differences between different subject groups in images acquired by the same modality. The first application requires identification and delineation ("segmentation") of distinct areas and landmarks in the brain. A procedure based on dynamic clustering and region growing algorithms has been developed that segments T1-weighted MR images into regions of CSF, gray and white brain matter. Its application to MR images of alcoholic and normal subjects yielded results consistent with a subjective segmentation. Automated methods for excluding extracranial tissues still need to be developed. For the second (within-modality) application, the gray-level information itself can be employed for image registration without the need for segmentation. A multiscale registration procedure has been developed that determines parameters of a general 3-D affine transformation (translation, rotation about an arbitrary center, anisotopic scaling and skewing) between volumes to be registered that minimizes the average squared gray-level difference between corresponding voxels. Successful registration of PET images achieving homogeneous registration variance across the entire brain section has been achieved for both within and between subject analyses. The method is also used for the co-registration of long time series of volumes acquired in functional MRI studies. An extension of this method is being developed based on using the cross-entropy between different volumes as a similarity measure instead of the gray-level difference. The cross-entropy approach will enable cross-modality registration.